Impact of Terrorist Attacks on Stock Market Volatility in Emerging Markets

Ayman Mnasri, Salem Nechi

PII: S1566-0141(16)30041-3
DOI: doi: 10.1016/j.ememar.2016.08.002
Reference: EMEMAR 460

To appear in: Emerging Markets Review

Received date: 7 April 2016

Revised date: 16 June 2016
Accepted date: 15 August 2016

Please cite this article as: Mnasri, Ayman, Nechi, Salem, Impact of Terrorist Attacks
on Stock Market Volatility in Emerging Markets, Emerging Markets Review (2016), doi:
10.1016/j.ememar.2016.08.002

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Impact of Terrorist Attacks on Stock Market Volatility in Emerging Markets

Ayman Mnasri1

College of Business and Economic, Qatar University

PT
Salem Nechi2

College of Business and Economic, Qatar University

RI
SC
Abstract

We use an event study methodology alongside an improved bootstrapping test to evaluate the

NU
impact of terrorist attacks on the volatility of stock markets in 12 MENA countries, and test for
regional financial integration. Results show that the impact of terrorist attacks on financial
markets‟ volatility lasts about 20 trading days, which is considered to be long compared to the
MA
term effect of similar events in developed markets. Moreover, we find evidence of regional
financial integration. Our robustness check shows that the Bootstrapping approach is more
robust, and that theoretical p-values might be misleading if underlying assumptions are violated.
D

JEL classification: G12; G14; G15; C12

TE

Key words: stock market volatility, emerging markets, event study, bootstrapping
P
CE
AC

1
Assistant Professor of Economics, Department of Finance and Economics, College of Business and Economics,
Qatar University, Doha, Qatar, P.Box. 2713. amnasri@qu.edu.qa, +974 44036756.
2
Corresponding author. Assistant Professor of Economics, Department of Finance and Economics, College of
Business and Economics, Qatar University, Doha, Qatar, P.Box. 2713. snechi@qu.edu.qa, +974 44036497.

1
 ACCEPTED MANUSCRIPT

1. Introduction
Most research examining the reaction of financial markets to terrorist attacks suggests that the
impact of these events on stock markets is limited and lasts for a short period (Essaddam and
Mnasri, 2015; Baumert et al., 2013; Chesney et al., 2011; Brounrn and Derwall, 2010; Arin et al,

PT
2008; Chen and Siems, 2004; among others). These conclusions were based on studies
examining the effects of terrorist attacks on stock returns in developed markets, where financial

RI
systems and institutions are well-established (Essaddam and Karagianis, 2014; Kollias et al.,
2011; Karolyi and Martell, 2010; Johnston and Nedelescu, 2005; among others). The current

SC
study contributes to the body of research on this subject in two ways. First, we investigate the
impact of terrorist attacks on market volatility, instead of the market returns. Second, we
examine the impact of terrorist attacks on the volatility of stock markets in emerging markets,

NU
instead of developed markets.

Our argument for the first contribution is that following a terrorist attack (or any negative shock),
MA
stock returns are usually expected to decrease and, depending on the severity of the shock, the
reaction of the returns might be dramatic. What would be interesting to know is the impact of
these attacks on stocks‟ volatility, instead. In fact, what is important for investors is the amount
of risk the market is subject to. The second contribution is related to our focus on the reaction of
D

stock markets in emerging markets. Our focus is justified by the fact that developed countries
TE

have well-established financial markets and strong institutions. Following terrorist attacks,
governments in these countries react quickly with a set of measures to smooth out the shocks,
while efficient economic and financial institutions (endowed with the appropriate tools and
P

resources) implement such policies and measures. In the case of developing countries, however,
CE

institutions are usually weak and the financial markets are unstable which may result in over
reaction of the markets and inefficient implementation of the policies and measures undertaken
by the government after terrorist attacks.
AC

To validate our contribution and fulfill our mandate, we investigate the impact of terrorist attacks
on stock market volatility in eleven (12) emerging countries from the Middle East and North
Africa (MENA) region. Our sample selection is motivated by four rationales. First, to the
authors‟ knowledge, no previous studies have investigated the impact of terrorist attacks on stock
market volatility in these countries. Second, given the strong political and economic ties between
many countries in this region and many developed countries (driven by the abundance of
hydrocarbon resources in this region), terrorist attacks induce reactions not only by local
governments and regional authorities but also by the international community. This fact may
delay the stock markets‟ adjustment following a terrorist attack. Third, most countries in our
sample lack economic diversification, which could make the impact of the terrorist attacks
magnified and longer lasting. Fourth, given that MENA countries share many cultural and
historical aspects, the reactions of the regional markets to a major attack occurring in one country
might be an indicator of the degree of MENA markets‟ integration.

2
 ACCEPTED MANUSCRIPT

Following a significant trend in the existent literature, we use an event study methodology to
investigate the impact of terrorist attacks on the volatility of stock markets. The advantage of
using this approach is to allow for measuring both, the size effect and the term effect of the
terrorist attacks on stock market volatility. The basic idea consists in estimating the volatility
processes in stock markets before the attacks, predicting the volatility levels for the post-events

PT
(attacks) period based on the assumption that no event had taken place, and then comparing the
predicted volatilities with the actual (abnormal) ones induced by the terrorist attack. In addition,

RI
we use the event study methodology to assess the degree of financial integration in the MENA
region; an application which we consider original in this literature.

SC
To test the significance of the impact of an event on the value of a variable, event study models
usually refer to the classical parametric testing approach (i.e., theoretical p-values). This

NU
approach may not be the best way to test the significance of the impact of terrorist attacks on
stock markets. In fact, one common issue in financial data is non-normality and serial
correlation. Parametric tests assume that the errors are normally distributed and are not serially
MA
correlated. When these conditions are violated, the test will be biased and the conclusions drawn
from it might be misleading. To overcome these drawbacks, some studies have considered
bootstrapping techniques to test the significance of some events on stock markets. For instance,
Białkowski et al. (2008) investigate the impact of national elections on stock market volatility in
D

27 OECD countries using an event study methodology alongside a bootstrapping test. In this
TE

paper, we propose an improved semi-parametric bootstrapping technique, and we show that it is

more efficient and more consistent than Białkowski et al. (2008) bootstrapping methodology.
P

The originality and the efficiency of the bootstrapping technique we use, represent an additional
CE

contribution. In fact, our bootstrapping algorithm overcomes the drawbacks of Białkowski et al.
(2008) bootstrapping method, which consists of randomly drawing (with replacement) N
country/date combinations from the entire set of available countries and dates, to match the
AC

number of events. As we will elaborate in more details in section 6.2, the test empirical
distribution obtained from their bootstrapping technique may not be consistent with the null
hypothesis. However, in our paper, the empirical distribution of the Cumulative Abnormal
Volatility (CAV) test is derived from re-sampled and rescaled residuals obtained from the
estimations of the benchmark GARCH equations before events. Hence, the implied test empirical
distribution is going to be perfectly consistent with the null hypothesis, and circumvent the
potential problems caused by the cross-sectional dependence or autocorrelation of the regression
residuals.

Our results show that terrorist attacks have a significant impact on stock markets‟ volatility in the
MENA region. This impact lasts for 20 trading days at least. Compared to the results of similar
studies for developed markets, this term effect is considered to be long, which could be attributed
to the lack of strong economic and financial institutions in the emerging markets examined.
Moreover, our investigation of the impact of the 2005 terrorist attack that targeted the former
Prime Minister of Lebanon, Rafik Hariri, on the volatility of stock markets in the region reveals
3
 ACCEPTED MANUSCRIPT

that regional markets did react to the attack in Lebanon, yet with a delay of about 10 days, and it
lasted for four weeks, suggesting that MENA markets are integrated, indeed. Finally, our
improved bootstrapping test shows not only that the semi-parametric test is more robust but also
that parametric tests might be misleading if the underlying assumptions are not met.

PT
The remaining of the paper is organized as follows. A brief review of the literature is outlined in
Section 2. The event study methodology and the approaches used to evaluate it are detailed in
Section 3. Section 4 describes the data and provides some preliminaries. The results and their

RI
discussion are presented in Sections 5 and 6, respectively. Section 7 concludes the paper.

SC
2. Review of Related Literature
The impact of political, economic, social, environmental and demographic events on stock

NU
markets has been extensively discussed in the literature. In fact, previous studies show that there
is evidence indicating that events such as wars, armed conflicts, elections, environmental and
non-environmental accidents, monetary policy, energy crises, etc. often exert considerable
MA
influence on markets‟ behavior. Nevertheless, the extent and duration of this impact may vary
significantly from one event to another and from one country to another. (e.g Choudhry, 1995;
Frey and Kucher, 2000, 2001; Amihud and Wohl, 2004; Schneider and Troeger, 2006;
D

Athanassiou et al., 2006; Bialkowski et al., 2008; Kollias et al. 2010; Guidolin and La Ferrara,
2010; Scholtens and Boersen 2011; Wang and Mayes 2012; Imbierowicz and Wahrenburg, 2013;
TE

Fiordelisi et al. 2014; Smales, 2014; Carpentier and Suret, 2015; Haitsma et al., 2016;
Kenjegaliev et al., 2016; Günster et al., 2016).
P

The increasing frequency of terrorist events worldwide has resulted in a growing demand for
CE

research investigating the economic consequences of terrorism. For instance, Estrada and
Koutronas (2016) estimate the regional impact of terrorist attacks using the Economic Impact of
AC

Terrorist Attack (EITA) model. According to this EITA model, the likelihood and the magnitude
of the attacks are associated with economic dynamics of the corresponding region. Applying
their analysis to recent terrorist attacks of Paris November 2015 and Brussels March 2016, the
EITA model suggests that the nature of the attacks, the economic resilience of French and
Belgium economies, and the security levels in both countries are important determinant of the
severity of the consequences and the contingency.

Given the importance of financial markets to the overall performance of the economy, a
significant trend in the literature of investigating the impact of these phenomena on stock
markets has emerged. The existing literature on the impact of terrorist attacks on stock markets
can be sorted into two categories: impacts on stock market returns and impacts on stock market
volatility. However, as we discuss below, the focus of both trends is on the impact of these
events on stock markets in developed markets, and on the degree of integration of these markets
based on their reactions. Moreover, to the authors‟ knowledge, none of previous studies focused
on the impact of terrorist attacks on stock markets in emerging economies and the MENA

4
 ACCEPTED MANUSCRIPT

countries in particular, and how regional markets react to these tragic events. In this section, we
review the literature on the impact of terrorist attacks on stock markets returns and volatility as
well as the studies relevant to the MENA region.

2.1 Impact of terrorist attacks on stock market returns

PT
The reactions of stock returns in international markets to terrorist attacks have been investigated
through assessing the effects of these attacks on market indices, sectorial indices, industrial

RI
indices or firms‟ performance. Hobbs et al. (2016) examine the effect on stock returns of 28
terrorist and military events occurring between 1963 and 2012. For each event, they calculate

SC
both the equally weighted and the value-weighted returns for all stocks included in the Center for
Research in Security Prices (CRSP) database for the day of the event, and the day immediately
following the event. The results show that stocks perform significantly worse on the days of

NU
terrorist events than on the days of military events, but the opposite is true for the day after.
Aloui and Nguyen (2014) combine the global Hurst exponent and Morlet wavelet multi-
MA
resolution analysis (MRA) to investigate the dynamic behavior of six selected stock markets in
the Mediterranean countries (including three from the MENA region, Egypt, Tunisia and
Turkey). According to their model, major extreme shocks, such as the 9/11 terrorist attack, have
effects on stock returns. They also find evidence of presence of long-term dependencies in stock
D

returns of all the considered markets, except for France.

TE

Although many models and approaches have been considered, the event study methodology has
been used the most. For instance, Aslam et al. (2016) use an event study methodology to
P

investigate the impact of terrorist attacks on stock returns in five Asian countries. The results
CE

suggest that while these events have significant impact on stock returns, the magnitude of these
affects varies with respect to country, attack type and severity of the attack. Similarly, using an
event study methodology, Aslam and Kang (2015) find that while terrorist attacks adversely
AC

affect Pakistani stock market, the impact is limited. They also find that the magnitude of this
impact is positively related to the severity of the attack. Chen and Siems (2004), Chesney et al.
(2011), Al-Ississ (2010), and Brounrn and Derwall (2010), among others, use the event study
approach to assess the effect of terrorism on international equity markets. Other methodologies
used include stylized macroeconomic models of the world economy (Abadie and Gardeazabal,
2008), world Capital Asset Pricing Model (Drakos, 2010) and tests for causality effects running
from terrorist activities to stock returns (Arin et al., 2008). Overall, the effects of terrorist attacks
on returns in international stock markets are found to be moderate and last for only short period
of time, as markets usually recover within days after the attack. However, these effects are
magnified when countries are integrated with global markets.

The impact of the three major terrorist attacks of 9/11/2001 (New York, US), 3/11/2004 (Madrid,
Spain) and 7/7/2005 (London, UK) on stock market returns have received particular attention in
the literature. One rationale behind this focus is to investigate the degree of integration of the
major international stock markets. Many methods have been adopted to study this issue.

5
 ACCEPTED MANUSCRIPT

Johnston and Nedelescu (2005), and Kollias et al. (2011) use an event study methodology to
examine the effects of these events on stock returns in global markets, whereas Nikkinen and
Vahamaa (2010) analyze the behavior of expected probability density functions around the three
attacks to capture the stock markets‟ sentiment. Hon et al. (2004) test for the contagion effect of
the 9/11 terrorist attacks using a bias-correlation procedure. Baumert et al. (2013) perform a

PT
comparative analysis between the impact of the Boston Marathon bombing (April 2013) on stock
returns in major international markets and the impact of the US 9/11 attacks.

RI
Overall, the main findings of these studies suggest that stock returns in all investigated markets
were negatively affected by the attacks, yet the extent and duration of this impact vary across

SC
countries. Broadly speaking, the reactions of well-developed markets to the attacks are smoother
and diminish more quickly. This implies that investors in these markets have learned how to

NU
assess the impact of these attacks objectively and not to overreact. Moreover, many studies
conclude that the terrorist attacks after 9/11 led to an increase in correlation across global
financial markets, suggesting that the benefits of international diversification in times of crisis
MA
are substantially reduced.

Other papers have also investigated the effects of terrorist attacks on companies‟ share prices.
For instance, Kolaric and Schiereck (2016) use an event study methodology to analyze the
D

impact of Paris 13/11 and Brussels 22/03 terrorist attacks on stock prices of 27 European and
TE

North American airline companies. Their results suggest that both events had strong short term
effect on the valuation of the of the airline companies in the sample. Using an event study
methodology, Apergis and Apergis (2016) investigate the impact of Paris 13/11 terrorist attacks
P

on the major companies of the defense industry worldwide. The upward trend in cumulative
CE

abnormal returns suggests that terrorist attacks have positive effect on stock returns of companies
of the defense industry. (See also Drakos, 2004; Choudhry, 2005; Karolyi and Martell, 2010;
Abadie and Gardeazabal, 2003; among others). Overall, the results show that terrorism
AC

negatively affects stock prices of publicly listed firms but the effects are rather small and last for
a limited time. Moreover, the impact is higher on firms operating in developed and more
democratic countries.

2.2 Impact of terrorist attacks on stock market volatility

While the impact of terrorist attacks (or other events) on stock returns has received considerable
attention in the literature, very few studies have tried to investigate the impact of terrorist attacks
on stock market volatility. For instance, Bentes (2016) adopts Fractionally Integrated GARCH
(FIGARCH) model to examine the impact of four major world crises, including the 9/11 terrorist
attacks, on the long memory behavior in the volatility of gold returns using daily data for the
period 1985–2009. She finds evidence of long memory in the conditional variance over the
whole sample period, but mixed evidence when considering sub-sample analysis. Essaddam and
Karagianis (2014) use a volatility event study approach to examine the impact of 44 terrorist
attacks on the stock return volatility of American firms operating inside and/or outside the US

6
 ACCEPTED MANUSCRIPT

over the period 1995–2010. Their results show that impact of the terrorist attacks on American
firms stock return volatility lasts for at least 15 days. They also find that firms located in rich
and/or more democratic countries experience a higher change in the volatility of their stock
returns compared to firms located in developing countries. Using a set of 28 events (terrorist
attacks) that happened in 17 countries (of these, only Egypt is in the MENA region), Essaddam

PT
and Mnasri (2015) refined the evaluation method of Essaddam and Karagianis (2014) by
proposing a bootstrapping method. Their results show that the abnormal volatility starts on the

RI
day of the attack and remains significant for a number of days.

2.3 The MENA region: Impact of Extreme shocks on stock markets and Financial Integration

SC
Eldomiaty et al. (2015) use fixed and random effects tests to evaluate the empirical associations
between the institutional quality and economic freedom in one hand and the stock market

NU
volatility in MENA countries in the other hand. The results show that countries with relatively
stronger institutions are characterized with less volatile stock markets. Moreover, stock market
MA
volatility can be mitigated and reduced when economic freedom is associated with an effective
enforcement of law and efficient regulations. Chau et al. (2014) examine the volatility of major
stock markets in the MENA region following the regional political unrest caused by the „Arab
Spring‟. Their results suggest that while the volatility of the Islamic indices increased
D

significantly in the aftermath of the events, the changes in the volatility of conventional indices
TE

was not significant.

Using granger causality tests and impulse response functions, Neaime (2016) examines
P

contagion vulnerability and the international and regional financial integration of MENA stock
CE

markets. His findings suggest that while GCC countries lack strong linkages with regional and
international markets, the other MENA markets have shown maturity and became financially
integrated with global financial markets and to lesser extend with regional markets. Bouri and
AC

Yahchouchi (2014) and Bouri and Azzi (2014) use multivariate model to examine the dynamic
relationship across the stock market returns in MENA, and Arab MENA countries, respectively,
between June 2005 and January 2012. They also investigate the impact of the 2008 global
financial crisis on the conditional variance and correlation of stock returns in the MENA
markets, and Arab MENA markets, respectively. The key finding of these two studies is that
MENA markets and Arab MENA markets are interconnected by their volatilities and not by their
returns.

None of the previous studies investigated the co-movements of MENA stock markets caused by
terrorist attacks. Moreover, to the authors‟ knowledge, none of the previous studies used the
event study methodology to examine regional integration of MENA stock markets. Given the
increased frequency of terrorist attacks in this region, we would expect a number of studies to
investigate the effects of such events on the returns and/or volatility of stock markets. The
current study is one step to explore this route. In particular, we examine the impact of terrorist

7
 ACCEPTED MANUSCRIPT

attacks on stock market volatility in the MENA countries using an event study methodology and
an improved bootstrap technique. The remaining work in this paper focuses on this question.

3. Methodology

PT
In this section, we discuss the methodology we use to measure the impact of terrorist attacks on
stock market volatility. The first step in our analysis is the presentation of the volatility event
study approach used to calculate the cumulative abnormal volatility (CAV), which is needed to

RI
test the impact of terrorist attacks on the volatility of stock markets. We then discuss the
parametric (based on a theoretical test) and semi-parametric (based on a bootstrapping test)

SC
inference techniques that we use to evaluate the impact of terrorist attacks on the volatility of
stock markets.

NU
3.1 Event Study Methodology

The impact of the terrorist attacks on the volatility of the stock markets is investigated using an
MA
event study approach. Kothari and Warner (2006) provide an extensive discussion of event
studies, their modeling, estimation and criticisms. The discussion presented here has benefited
from the work by Bialkowski et al. (2008) and the use of notation is kept to a minimum. We use
a generalized autoregressive conditional heteroscedasticity GARCH(1,1) model to capture the
D

characteristics of the second moment of stock returns‟ distribution.3 The specification of the
TE

GARCH(1,1) model is given as follows:

(1)
P
CE

(2)

where stands for the returns index for country i at time t, is the global stock market index
AC

at time , the error term denotes country i‟s specific component of its own index returns and
is the conditional variance of the error term.

Following the literature, we estimate Equations (1) and (2) simultaneously using the maximum
likelihood method over the period immediately preceding the event. Note that the period
preceding the event window shouldn‟t include any other (similar) event(s) to avoid overlapping
effects. Brown and Warner (1985) estimated a similar model using a period of 250 daily returns
immediately before the event (which represent one year). This period selection became a
benchmark in the literature. However, for our purposes, we use a window of 500 daily returns
instead. The reason for considering an extended period is that 250 observations of daily returns
might not be enough to estimate a GARCH processes properly. In fact, longer periods, over
which the GARCH process is estimated, are usually associated with more accuracy.

3
Many studies have shown that GARCH models can capture most of the common characteristics of stock returns,
especially daily returns. See ,for example, Bollerslev (1987).

8
 ACCEPTED MANUSCRIPT

Note, however, that very long periods of estimation may cause problems in some cases. In fact, if
the events occur frequently, the periods of estimation and the effects of the events will overlap,
which may result in misleading conclusions and interpretations. One way to deal with such
problems is by eliminating events that are relatively close; however, some of them might be very
important. Given that terrorist attacks have high frequency in the MENA region, our model will

PT
face a similar issue if the period of estimation is very long. Our choice of a period of estimation
of 500 daily returns immediately preceding the event window, in addition to being consistent

RI
with a trend in the financial literature, has a theoretical and practical foundation: it will allow us
to keep the most significant terrorist attacks in our sample.4

SC
The GARCH model serves us in comparing the abnormal volatility of the error term, , around
the event date in relation to the same term in non-event circumstances. In fact, the GARCH

NU
model which we estimated using a window of 500 daily returns immediately before the terrorist
attack is used to predict the volatility of the returns index if no attack happened. It is important to
note, however, that the conditional volatility estimated by Equation (2) is only a one-step-ahead
MA
forecast and cannot be used to estimate the volatility of the subsequent periods beyond the day of
the event. In other words, the error term of the day of the terrorist attack ( ), , has an
important impact on the volatility of the index returns of the days after the event. For this reason,
D

the volatility at any point in time after the terrorist attack (i.e., for any ) is forecasted on the
basis of the information obtained immediately before the event. Accordingly, to forecast the
TE

conditional volatility for the day of the event window (i.e., the day after the terrorist
attack or k steps ahead), we use the information set, , which is available on the last day of the
P

estimation window :
CE

(3)
AC

The distribution of the residuals during the event window is given by

, where is the abnormal returns generated at time t (after the event, ) and
is the multiplicative effect of the event on volatility. One fundamental assumption in event
study models is that the events are completely unpredictable. The null hypothesis in our model is
that the volatility of the stock markets is not affected by terrorist attacks. If the null hypothesis is
true, will equal one. This parameter will be of primary interest in our analysis.

The next step in our methodology is to de-mean the residuals using the cross-section average and
to calculate their variance under the assumption of residual orthogonality. Doing so gives us
residuals that are normally distributed with a mean of zero and the variance:

(4)

4
See, for example, Hwang and Valls Pereira (2006).

9
 ACCEPTED MANUSCRIPT

where is the event-independent de-meaned residual variance and N is the number of

events (i.e., terrorist attacks) included in our sample.

Our estimation of the parameter is based on the methodology of Boehmer et al. (1991) and
Hilliard and Savickas (2002), which combines residual standardization with a cross-sectional

PT
approach. Accordingly, the estimate is the cross-sectional variance of the de-meaned

residuals, standardized by :

RI
, (5)

SC
where the residual is obtained from Model (1): and

NU
When (the null hypothesis is true), the de-meaned standardized residuals follow a
standard normal distribution. Therefore, the abnormal percentage change in volatility on any day
MA
t of the event window is given by . More generally, the CAV over a given event
window is calculated as follows:

. (6)
D

Having defined the parameter we wish to estimate and its distribution, we can now proceed with
TE

hypothesis testing. Under the null hypothesis, terrorist attacks have no effect on stock market
volatility. Following the discussion above, this can be formulated as shown below:
P

(7)
CE

This is equivalent to:

AC

(8)

Given that under the null hypothesis, is a variance of N independent random variables
following a standard normal distribution, , the statistic follows a
distribution with (N – 1) degrees of freedom (i.e. ).

Thus, the test statistic for the current null hypothesis is:

. (9)

3.2 Testing the Approaches

To conduct this test, standard inferential models (parametric) assume explicitly that the
underlying assumptions of the theoretical tests are satisfied. In other words, the error terms, ,
are normally distributed and are not serially correlated. If any of these assumptions is violated,
the test statistic is biased and the results could be misleading. Efron (1979) introduced a semi-

10
 ACCEPTED MANUSCRIPT

parametric approach for addressing this issue. Efron‟s approach consists of the bootstrap
methodology in which normality is not a crucial assumption and serial correlation is dealt with.
Based on this technique, the empirical distribution of the CAVs is simulated under the null
hypothesis and is compared with the CAV in the period following the terrorist attack.

PT
The idea behind bootstrapping is relatively simple yet very powerful. It provides a popular way
to perform inference that is more reliable in finite samples than inferences based on conventional
asymptotic theory. In other words, bootstrapping consists in generating a large number of

RI
“bootstrap samples” that resemble the real data under the null hypothesis that will be used to
approximate the cumulative distribution function of the test statistics and its p-value. Bootstrap

SC
methods have recently become very popular because of the significant decrease in the numerical
computation cost, thanks to the dramatic increase in the speed of computers. Beran (1986), and

NU
Hall and Titterington (1989) were among the first papers to study bootstrap testing formally.

In certain circumstances, bootstrap tests are exact. In other words, the probability of rejecting the
null at level α is exactly α. This is the case when the distribution of the test statistic does not
MA
depend on unknown parameters (i.e. it is pivotal). Even if the test statistic is not pivotal,
bootstrapping will often yield tests that are very close to being exact. Actually, when the test
statistics are too complicated or do not have a known asymptotic distribution (or a non-standard
D

asymptotically distribution), bootstrapping provides a significantly better inference than the

TE

standard asymptotic tests do. Indeed, Horowitz (1994) and Hall and Horowitz (1996), among
others, have found that bootstrap tests are generally much more reliable than asymptotic tests.
Hall (1992) argues that bootstrapping can reduce the error in the rejection probability by a factor
P

of , or even more. Davidson and MacKinnon (1999) reported similar results.

CE

When the distribution of the observed test statistic is unknown, the usual approach is to replace it
by the asymptotic cumulative distribution function. However, when the number of observations
AC

is not high enough, the asymptotic distribution may not be a good approximation of the test
statistic finite distribution. In this case, the test statistic‟s empirical distribution function implied
by bootstrapping is believed to provide a much better approximation and more accurate p-values.

In this paper, our main concern is to investigate the abnormal volatility, if any, of the stock
returns after terrorist attacks. To do so, we make use of the well-known CAV statistical test.
Intuitively, this test compares the magnitude of the residuals computed over the window event
(i.e. the period over which we would like to test the abnormal volatility) with the predicted
volatility when the null hypothesis is true. In fact, we estimate the N GARCH (1,1) equations
using only data BEFORE the event. Hence, the test statistic (CAV) will depend exclusively on
the residuals computed after the event. According to the null hypothesis, the distributions of
these residuals should still belong to same set of distributions where the variance is governed by
the same process. Hence, according to the null hypothesis, there is no systematic increase in the
volatility of the stock returns after the event date (abnormal volatility). Hence, implementing the
bootstrap test does not require us to estimate the GARCH equations for every bootstrap sample,

11
 ACCEPTED MANUSCRIPT

since the statistic test depends on empirical residuals, which are simply the difference between
the real stock returns and the predicted ones (which are derived from parameters that were
already estimated before the event). The bootstrap data generation process in our case consists of
re-sampling the rescaled residuals obtained from the estimations of the GARCH equations before
the event. In fact, when adjusted (i.e. when they have the correct variance under the null

PT
hypothesis), the empirical test statistic‟s distribution is the best approximation of its finite
distribution (under the null hypothesis).

RI
The conventional CAV test assumes that these residuals are normally distributed. However, it is
hard to believe that stock return shocks are Gaussian. Indeed, according to our results, the

SC
bootstrap p-values seem to be higher than the theoretical ones. This means that bootstrapping
tests give a higher rejection rate of the null hypothesis, which, in turn, implies that the

NU
bootstrapping evaluation method tends to capture the impact of the terrorist attacks on stock
market volatility more than the conventional CAV testing method. One potential explanation is
the relatively high excess kurtosis (and/or high skewness) of the finite distribution of the test
MA
statistic compared to the normal distribution (excess kurtosis=0). Hence, „extreme‟ values have
significantly higher probabilities than the case of a Gaussian distribution (flatter distribution). It
is not surprising that in some cases, we can have non-significant bootstrap p-values while the
theoretical p-values are significant. In this sense, we can qualify our bootstrap method as a semi-
D

parametric bootstrap (or non-parametric bootstrap), since we are bootstrapping residuals (re-
TE

sampling with replacement).

Another strong assumption of the theoretical CAV test is the absence of any serial correlation in
P

the residuals. In fact, when the event window is longer than one day, the CAV test includes
CE

consecutive residuals of the same country‟s equation. It is hard to believe that those residuals are
serially uncorrelated. To deal with this issue, we choose the residuals in a chronologically
consecutive manner every time we generate a bootstrap sample. Hence, our bootstrap method is
AC

comparable to a moving block bootstrap.

Bootstrapping algorithm:

The procedure for generating residuals and the distribution to be used to compute the bootstrap
p-value is illustrated below:

 For each of the N terrorist attacks (or countries5), we estimate the GARCH (1, 1) equation
using the 500 observations that immediately precede the window event (the window event
is the period following the attack over which we want to test the impact on volatility).

5
N indicates the number of countries when we investigate the impact of only one attack on the stock markets of
other countries (Section 5.3).

12
 ACCEPTED MANUSCRIPT

 Residuals of the GARCH estimations are sorted into a (500*N) matrix called . Each
vector of this matrix, , consists in 500 residuals of Equation (attack or event .

 Create a new matrix by rescaling matrix so that for each vector , its elements have

PT
mean of zero and a variance of 1.

RI
 Multiply each vector (of matrix ) by the corresponding predicted conditional standard

SC
deviation (under the null hypothesis). The result is N matrices , each of which has
( ) vectors. ) also corresponds to the number of days in the event
window.

NU
Consequently, each attack is associated with a (500*( ) matrix, , where the
adjusted residuals of each vector have a mean of zero and the variance , where s =
MA
. It should be noted, however, that the elements of these vectors will have the same
distribution as the original residuals, which is, in turn, assumed to be the same after the event
under the null hypothesis.
D

Our bootstrapping algorithm has 10,000 iterations and consists of the following steps:
TE

1. For each event we quasi-randomly assign ( ) elements from so

that the first element belongs to the first vector and so on. In order to capture any
P

potential systematic autocorrelation in the original distribution of the residuals, we

CE

choose the ( ) elements in a chronologically consecutive manner.

2. Using these rescaled and re-sampled residuals, we compute the CAV according to the
AC

formula and we then compare it to its theoretical counterpart already calculated in the
previous section.

3. We go back to Step 1.

4. Once all iterations are done, we calculate the p-value according to the standard
bootstrap method.

We usually use bootstrapping techniques when the theoretical distribution of a given statistical
test is unknown or too complicated. It is also recommended to use bootstrapping when we
believe that the assumptions we used to identify the theoretical distribution of the test, are not
very realistic or can be easily violated. In our case, the CAV is, theoretically, supposed to follow
a chi-squared distribution, which is the sum of the squares of k independent standard normal
random variables. Thus we can suspect the normality and the serial independence assumptions of
the residuals. Indeed, it is very unlikely that the residuals of the daily returns yield rates are

13
 ACCEPTED MANUSCRIPT

serially uncorrelated. Moreover, the size of the sample, N (the number of terrorist attacks), is
considered to be small.

The originality of this bootstrapping strategy comes from the fact that it takes all of these
considerations into account. In fact, we used the distribution of the residuals estimated before the

PT
event to compute the CAV of the period following the same event. Hence, under the null
hypothesis, this strategy is perfectly consistent and efficient since, in this case, the distribution of
the residuals should be the same before and after the event. Choosing the residuals to be

RI
chronologically consecutive allows us to capture any potential structural autocorrelation in the
daily yield rates of different indices, which can influence the empirical distribution of the

SC
statistical test and consequently presents a very important source of bias.

4. Data and preliminaries

NU
As emphasized in the introduction, our primary focus is on emerging markets. Our data include
13 stock markets (12 countries) from the MENA region. Although the MENA region includes
MA
more countries, we excluded them for various reasons. We excluded Iraq, Syria, Libya and
Yemen because these countries have been experiencing civil wars for the last 5 years (at least)
and a terrorist attack could be interpreted as an expected event. Moreover, given the lack of data
D

availability for Algeria and Iran, the impact of terrorist attacks on stock market volatility in these
two countries was not investigated. However, we include the terrorist attacks that happened in
TE

these two countries in our event sample because we believe that they are big regional economies
and dramatic events in them may affect other regional economies. We collected data for the
P

market indices for the remaining countries in the region over the period 2000–2015. The markets
CE

under consideration are the seven stock exchange markets of the GCC (Bahrain, Kuwait, Qatar,
Saudi Arabia, Oman, Dubai and Abu Dhabi), three stock exchange markets in the Middle East
(Jordan, Lebanon and Turkey) and three exchange markets from North Africa (Egypt, Morocco
AC

and Tunisia). Note that for Egypt, we did not consider any event after June 2013, as the country
entered a period of widespread violence after the military coup that took place early July 2013,
so any terrorist attack after that date would be an expected event. For Tunisia, we extended our
dataset to April 1998 to satisfy the window of 500 daily returns.

Terrorist attacks have been gleaned from the Global Terrorism Database. The Global Terrorism
Database documents all incidents of terrorism across the globe since 1970.6 Our criteria for
selecting the terrorist attacks (events) differ from one country to another. If a country is subject
to a relatively high frequency of terrorist attacks (Saudi Arabia and Turkey, for example), we
selected the „important‟ attacks only. An attack is considered „important‟ if the damage
(casualties and financial losses) it caused were significant and/or it targeted an important facility
(such as an oil or gas site, refinery plant, etc.) or a large city. On the other hand, for countries

6
The year 1993 is excluded from the Global Terrorism Database because of data issues related to that year. This
has no effect on our dataset, as we start well after that date.

14
 ACCEPTED MANUSCRIPT

that are relatively stable and have a very low frequency of attacks (such as Bahrain, Kuwait,
Qatar and the United Arab Emirates), we investigate the impact of any terrorist attack even it is
„small‟ with no serious material losses or casualties. The rationale for this approach to selecting
the events is that although „small‟ events have no significant effect (sometimes going unnoticed)
in countries exposed to frequent „big‟ attacks, similar events might be seen as a significant threat

PT
in countries unfamiliar that are with such events.

One common criterion related to the dates and intervals between events for all countries is worth

RI
detailing, though. As explained in the methodology, for any event to be selected, a window of
500 trading days (2 years) immediately before the day of the attack with no other events is

SC
required. If a potential/candidate event (even a very important one) was preceded by another
event (even if it was not very important) within the 500-day window, it will not be selected. By

NU
doing so, we avoid any bias in the estimated benchmark volatility, GARCH (1,1). Because a
large number of terrorist attacks in the MENA countries happened within an interval of less than
2 years, the application of this criterion had shortened list of attacks to be considered to only 20.
MA
Table 1 summarizes the attacks retained in our model.

5. Results
D

All our tables reporting the results of the event-related volatility study have the same notation,
which can be summarized as follows. The first column of each table refers to the event window
TE

in which the volatility is investigated. Each window has two entries: The first entry refers to the
starting date and the second entry refers to the end date. Moreover, all entries are benchmarked
P

against the day of the attack, denoted “1”. For example, a window of (1,3) corresponds to an
CE

investigation of the volatility between the day of the attack and third day after the attack, a
window of (1,10) refers to the period between the day of the attack and the 10th day after the
attack, a window of (6,10) refers to the period between the 6th and 10th day after the attack, and
AC

so on. On the other hand, a window of (-1,-5) corresponds to the five days immediately before
the terrorist attack and a window of (-15,-20) refers to the period between the 20th and 15th days
before the attack. Note that if the attack takes place on a day when the market was closed (a
weekend or national holiday, for example), “1” would refer to the first trading day following the
attack. More generally, the starting day of the window is and the last day of the window is .

15
 ACCEPTED MANUSCRIPT

Table 1: Terrorist Attacks – The Events

Event Number Country Date Description
1 Turkey 14/11/2003 Two car bombings on synagogues in Istanbul killed at least 20 people and injured 302 others

PT
2 Turkey 27/7/2008 Two bombs exploded minutes apart in one of Istanbul's busy shopping districts on the European side.
An explosives-laden vehicle was detonated outside City Hall in Reyhanli town, Hatay province. At least 53 civilians were killed and

RI
3 Turkey 11/5/2013
another 140 were injured across the two blasts.
4 Turkey 10/10/2015 Two bombs were detonated outside Ankara Central railway station, killing 102 people

SC
A suicide bomber detonated explosives at a wedding being held at the Radisson Hotel in Amman; 57 people were reported killed and
5 Jordan 9/11/2005
approximately 100 others were wounded.

NU
Unidentified perpetrators set off a car bomb near the motorcade of former Lebanese Prime Minister Rafik Hariri, killing the former
6 Lebanon 14/2/2005
leader and 11 others
An explosives-laden vehicle detonated in Achrafieh district, Beirut City. Deaths included Wissam Al Hasan, a senior officer in the Internal
7 Lebanon 19/10/2012

MA
Security Forces, linked to the anti-Damascus camp of 14 March
8 Egypt 7/10/2004 A truck bomb struck the Hilton Hotel in Taba, Egypt. There were 34 deaths and 159 wounded
In a series of five related incidents, suicide bombers blew themselves up; 45 people were killed as a result of these attacks (33 victims
9 Morocco 16/5/2003
and 12 suicide bombers)

ED
10 Morocco 11/3/2007 In a café in the slums of Casablanca, Morocco, suicide bomber Abdelfettah Rayid detonated an explosive attached to his person
On Thursday, in the Djemma el-Fna Square in Marrakech Medina, Marrakech-Tensift-El Haouz, Morocco, 17 people, mainly foreigners,
11 Morocco 28/4/2003

PT
were killed when Adil Al Athmani detonated two remote controlled improvised explosive devices in the Argana Café.
A truck filled with cooking oil smashed into the El Ghriba Jewish shrine on Djerba Island. At least nine people were killed in the
12 Tunisia 11/4/2002
explosion.
CE
More than 25 extremists engaged in a gunfight with authorities in Soliman, Tunisia. Twelve assailants were killed and another 15 were
13 Tunisia 3/1/2007
arrested.
AC

14 Tunisia 6/2/2013 Assailants opened fire on Chokri Belaid and killed him outside of his private residence in Tunis City.
15 Bahrain 3/3/3014 A remote-controlled explosive device was detonated near a police officer. The blast killed three officers and injured 11 other people
16 UAE 1/12/2014 A female assailant killed a US teacher in the Reem Island mall, Abu Dhabi City.
Saudi In a series of nearly simultaneous attacks, suicide bombers blew themselves in a truck loaded with explosives in the compound of Al-
17 12/5/2003
Arabia Jadawel. At least two persons were killed in the incident. Al-Qa’ida was suspected to be responsible for the incident.
Saudi ISIL claimed responsibility for a suicide bombing during midday prayers at a Shia mosque in eastern Saudi Arabia. The Saudi health
18 22/5/2015
Arabia ministry said that at least 21 people had been killed and more than 120 others injured.
An Egyptian suicide car bomber rammed his vehicle into the Doha Players Theatre in Doha, Qatar. The car bomb killed one British man,
19 Qatar 19/3/2005
Jonathan Adams and injured 12 other people of mixed nationality
A suicide bombing took place at a Shia mosque in Kuwait. ISIS claimed responsibility for the attack. Twenty-seven people were killed and
20 Kuwait 26/06/2015
227 people were wounded.
UAE, United Arab Emirates; ISIL, Islamic State of Iraq and the Levant; ISIS, Islamic State of Iraq and Syria

16
 ACCEPTED MANUSCRIPT

The second and third columns of each table correspond to the theoretical and bootstrap p-values,
respectively, while the last column shows the calculated CAV during the window under
consideration.

5.1 Terrorist Attacks and Breaks in Stock Market Volatility

PT
To validate our argument that terrorist attacks may affect stock markets‟ volatility, we test
whether volatility rose around these tragic events. To do so, we create a symmetric event

RI
window of 20 trading days around the date of each attack and calculate the CAV before and after
the event. That is, for a given attack, we consider 10 days before the attack and 10 days after the

SC
attack, and conduct our test based on the steps we described in the methodology section. Table 2
below displays the results of this exercise. For the 10 days before the terrorist attack, (-10,-1), the
null hypothesis cannot be rejected, as both the theoretical and bootstrap p-values are very high. It

NU
is interesting to note that a significant break in the CAV took place in the post-attack days of the
window, as the CAV increased substantially over the second half of the event window (1,10).
MA
This observation is supported by both tests, which confirm that the null hypothesis is rejected at
the 1% significance level. In other words, there is strong evidence that the volatility increased
after the terrorist attacks.
D

Next, we narrow the symmetric event window to include only the day immediately before the
terrorist attack, (-1,-1), and only the day after the terrorist attacks, (1,1). The reported results of
TE

this exercise (in Lines 3 and 4 of Table 2) confirm the findings of the symmetric event window
of 20 trading days, namely that terrorist attacks induced high volatility in the stock markets.
P
CE

Table 2: Terrorist Attacks and Cumulative Abnormal Volatility

(CAV) Breaks
AC

Window Theoretical Bootstrap

CAV
p-value p-value

(-10,-1) 0.954 0.737 -1.666

(1,10) 0.000 0.001 14.619

(-1,-1) 0.524 0.450 -0.054

(1,1) 0.000 0.012 2.150

These results show that in our sample countries, terrorist attacks clearly affect stock market
volatility. As outlined in the introduction and the literature review sections, previous studies have

17
 ACCEPTED MANUSCRIPT

shown that the effects of terrorist attacks on stock markets (returns or volatility) did not last long,
as the returns or volatility returned to the normal level within days. Those conclusions were
based on studies conducted in developed markets. Can a similar conclusion be drawn for our
sample of emerging markets? The next step in our analysis investigates this question.

PT
5.2 Term Effect of Terrorist Attacks

Our investigation of the duration of the impact of terrorist attacks on our sample of emerging

RI
markets is conducted in two ways. First, we test for the effect using event windows that have the
same starting date, which is the day of the attack (noted “1”), but where the end date is extended

SC
to investigate the cumulative impact of the attacks on the abnormal volatility. The second way of
investigating the impact of the attacks is to conduct different tests using partial event windows
related to periods (10 trading days) that do not overlap. More precisely, we conduct tests for the

NU
following windows: (1,10), (11,20) and (21,30). It is worth noting that all results reported in
Tables 3 and 4 are based on the same GARCH benchmark model estimated over the 500 trading
MA
days before the events. Table 3 below displays the results of the volatility event study for the
post-attack period.

Table 3: Cumulative Abnormal Volatility (CAV) After the Terrorist Attacks

D

Window
Theoretical p-value Bootstrap p-value CAV
TE

1,1 0.000 0.008 2.6504

P

1,2 0.000 0.000 8.983

CE

1,3 0.000 0.000 14.5887

AC

1,5 0.000 0.000 17.8761

1,10 0.000 0.000 22.5786

1,15 0.000 0.000 24.1716

1,20 0.000 0.000 34.5966

1,30 0.000 0.000 34.1611

For all windows, both the parametric and semi-parametric tests show clearly that terrorist attacks
are followed by increased volatility. A significant increase in the CAV is initiated the first day
after the event and keeps increasing steadily for almost three weeks (window [0,20]). The impact
becomes limited afterwards as the CAV stops increasing after 20 trading days, indicating that

18
 ACCEPTED MANUSCRIPT

markets settle down almost one month after the event. The last line of Table 3, the window
(1,30), demonstrates that after one month, the CAV starts to decrease.

Next, we investigate the effect of the terrorist attacks for different window events that have
different starting and end dates, but the same estimation window. This exercise will allow us to

PT
evaluate the impact within the specified window, which is isolated from the impact, if any, on
previous dates. In Table 4, we report the results for three event windows of 10 days each with
different starting days. Specifically, we test for increased volatility in the window of the first 10

RI
trading days following the terrorist attack (0,10), followed by a test for the second 10 trading
days (11,20) then a window of the third 10 trading days (21,30). The results are consistent with

SC
the findings of Table 3. Both the theoretical and bootstrap tests are conclusive and have
significantly small p-values for the first and second windows and p-values that are not significant

NU
for the last window. Again, this implies that the impact of the terrorist attacks lasts for 20 trading
days and diminishes afterwards.

Table 4: Cumulative Abnormal Volatility (CAV) After the

MA
Terrorist Attacks for Three 10 Trading Day Windows

Window Theoretical Bootstrap

CAV (n1,n2)
D

p-value p-value
TE

0,10 0 0 22.579

11,20 0 0.002 12.018

P

21,30 0.654 0.549 -0.435

CE
AC

The fact that the reaction of the markets is sustained for about a month (20 trading days) might
be explained by the fear of investors that subsequent attack(s) may occur. As we mentioned in
the data section, MENA countries are subject to many terrorist attacks, some of which are close
in time and space (same country). Moreover, over the last two decades, international political
reactions to terrorist attacks happening in the Middle East have been very strong and radical
(including military) and had led to increased tension in the region. Aware of these facts,
investors usually wait to see if subsequent attacks will follow and try to see the reactions of local
authorities and the international community as well, leading to persistence of the excessive
volatility.

5.3 Market Integration and Reactions to Terrorist Attacks

Many markets in the MENA region have many similarities and share common features and
characteristics, suggesting that they may be integrated. In fact, previous studies shown have that
some markets reacted to good or bad news in other regional markets. For instance, Abul Basher

19
 ACCEPTED MANUSCRIPT

et al. (2014) used copula models to investigate the dependence of GCC stock markets over the
period 2004–2013. They found evidence of conditional lower dependence among all markets that
was stronger than the upper tail dependence, suggesting that GCC markets react more
aggressively when bad shocks take place. Assif (2003) and Al-deehani and Moosa (2006) found
evidence of GCC market integration.7 Marashdeh (2005) uses an autoregressive distributed lag

PT
approach to investigate the integration of the stock markets of Egypt, Turkey, Jordan and
Morocco. His results suggest that these markets are integrated. Using EGARCH-M with a

RI
generalized error distribution, Yu and Hassan (2008) found evidence of large and predominantly
positive volatility spillover among MENA stock markets, suggesting integration.

SC
In the present study, we investigate the integration among MENA markets by examining the

NU
impact of a terrorist attack occurring in one country on the stock market volatility of other
regional markets. For our purpose, we select a terrorist attack which we consider to be a turning
MA
point in the region, as increased tension took place because of it, and serious economic and
political consequences lasted for a long time after it.
D

Table 5: The Lebanese Terrorist Attack and Regional

TE

Markets’ Reactions

Window Theoretical Bootstrap

P

CAV
p-value p-value
CE

1,5 0.232 0.346 0.882

AC

6,10 0.010 0.163 3.488

11,15 0.000 0.053 7.789

16,20 0.000 0.014 13.244

21,25 0.000 0.000 26.097

26,30 0.000 0.015 13.699

31,35 0.176 0.325 1.174

The event we selected is the 2005 terrorist attack that targeted the former Prime Minister of
Lebanon, Rafik Hariri. This tragic event led to political disturbance and also to economic and
7
A good survey of GCC market integration literature can be found in Abul Basher et al. (2014).

20
 ACCEPTED MANUSCRIPT

social disruption in the Middle East. Moreover, not only local or regional authorities reacted to
this event but also the international community (including the United Nations, and influential
countries such as the US, the UK and France) reacted with significant political, social and
economic measures.

PT
Since a major terrorist attack that disturbs economic activity and threatens social and political
stability can induce serious fears among investors, there is a potential risk that the capital
markets in neighboring countries will be affected, especially if the region‟s markets are

RI
integrated. In what follows, we investigate the impact, if any, of the Lebanese 2005 terrorist
attack on the volatility of stock markets in the region, namely the five GCC countries (six stock

SC
markets), Egypt, Jordan and Turkey. To do so, we benchmarked our GARCH model around the
terrorist attack on the former Prime Minister of Lebanon, which happened on Monday February

NU
14th, 2005. For each of the regional markets investigated, we selected the 500 trading days
immediately before that date, estimated the GARCH(1,1), calculated the CAV and conducted the
parametric and semi-parametric tests to check if a change in the volatility took place. We
MA
conducted the tests on a sequence of five trading day event windows. In other words, we verify
the impact of the attack during the first five days following the event, (1,5), then the impact of
the event on the volatility during the second 5 days, (6,10), and so on. Note that we excluded
Qatar because it experienced a terrorist attack before Lebanon‟s event, making it impossible to
D

get a 500 trading day window without any other attack. We also excluded North African
TE

countries (Tunisia and Morocco) because no evidence can be found in the literature of the
financial integration of these markets with Middle Eastern markets.
P

Table 5 displays the outcome of our investigation. For the first five days following the tragic
CE

event, there is no evidence of a reaction in the volatility of the stock markets we investigated. In
fact, neither the theoretical nor the semi-parametric test rejected the null hypothesis, implying
that the volatility in these markets was not affected. For the second event window (6,10), the
AC

CAV increased substantially (from 0.882 to 3.488) but only the parametric test was significant,
whereas the bootstrapping test implied no change in the volatility. Both the theoretical and
bootstrapping tests for the subsequent four event windows (11,15), (16,20), (21,25) and (26,30)
were significant, indicating that the regional markets reacted to the attack in Lebanon. The effect
became limited afterwards, as the CAV decreased during the window (31,35) and neither test
rejected the null hypothesis. Figure 1 below displays the evolution of the CAV for the windows
we considered.

21
 ACCEPTED MANUSCRIPT

Figure 1: Regional Cumulative Abnormal Volatility (CAV) Following the

Lebanese Terrorist Attack.

Cumulative Abnormal Volatility

PT
30.000

RI
25.000

SC
20.000

15.000

NU
10.000

5.000
MA
0.000
1,5 6,10 11,15 16,20 21,25 26,30 31,35

window-event
D
TE

As can be seen from Figure 1, although the regional markets reacted during the 30 trading days
following the attack, the impact was limited after 20 trading days and the markets settled down
P

within 30 days. The conclusion that can be made here is that regional markets did react to the
event that happened in Lebanon but with delays. The fact that the markets reacted to the event
CE

represents evidence that is supportive of previous findings in the literature, suggesting that many
MENA capital markets exhibit some integration. The delay in the reaction of regional markets
AC

might be attributed to the fact that investors across the region wait to see the measures and
actions taken by local authorities as well as the international community (as explained above). A
few days after the attack, which was not claimed by any organization or country, the intelligence
agencies of many countries started to point out some suspects (including local militia and
neighboring countries), leading to increased tension between several countries in the region,
which prepared for the worst scenarios. In the subsequent weeks, the effects of this event on
regional markets decreased.

The above findings suggest that emerging markets in the MENA region are indeed integrated.
Moreover, regional diversification seems to be compromised during periods of crisis.

5.4 Parametric vs. Semi-Parametric Tests: Robustness Check

As emphasized in the methodology section, our evaluation of the impact of terrorist attacks on
stock market volatility in emerging markets was based on two types of test. The difference
between the two tests is related to the calculations of the p-value. Whereas the calculation of the

22
 ACCEPTED MANUSCRIPT

first test (parametric) is performed under the assumption that the residuals satisfy the theoretical
assumptions (normally distributed and not serially correlated), the calculations of the second test
(semi-parametric) are conducted without any prior assumption of the residuals using
bootstrapping techniques. The problem with the first test is that potential complications may
arise if one or more of the underlying assumptions is not satisfied. In fact, many studies show

PT
that financial data is generally not normal. Unreported results for the normality and serial
correlation tests indicate that these two assumptions are violated. We used the Bai and Ng (2005)

RI
normality test and the Ljung and Box (1978) autocorrolation test. For all series, both tests
strongly reject the null hypothesis of normality and no serial correlation, respectively.

SC
Table 6: Robustness Check

Theoretical Bootstrap

NU
Window
CAV
p-value p-value

-30,-30 0.678 0.552 -0.174

MA
-30,-29 0.045 0.183 0.841

-30,-28 0.007 0.137 1.573

D

-30,-26 0.059 0.251 1.184

TE

-30,-21 0.468 0.497 0.048

P

-30,-16 0.022 0.276 2.636

CE

-30,-11 0.003 0.233 4.221

-30,-1 0.025 0.336 3.587

AC

In such circumstances, bootstrapping methodology becomes a useful additional statistical

approach to test our work. Bootstrap tests are proven to be robust as they help to resolve the
problems related to the violation of the assumptions related to the residuals. Our robustness
check was built using the following steps. We estimated our GARCH(1,1) model using 500
trading days but not immediately before the terrorist attack. Instead, we created a window of 30
trading days immediately before the event and estimated our GARCH(1,1) using a 500 trading
days window that ends one day before the newly created window of 30 days. In other words, the
new benchmark date for this exercise is 30 days before the actual event (-30), which does not
correspond to any terrorist attack. Next, we forecasted the volatility of the stock returns during
the 30 days following the new benchmark based on our GARCH(1,1). Obviously, we would not
expect any change in the volatility during the new window of 30 days prior to the actual attack.
To verify this claim, we conducted our regular theoretical and bootstrap tests to identify any

23
 ACCEPTED MANUSCRIPT

change in the volatility. Not very surprisingly, the outcome of the parametric test was counter-
intuitive. Table 6 reports the detailed results of different event windows, all of which have the
same starting date (the new benchmark, 30 days prior to the terrorist attack) but the end date
changes. For example, the event window (-30,-30) corresponds to the new benchmark day, (-30,-
21) corresponds to the 10-day event window starting at the new benchmark day and ending nine

PT
trading days later (i.e., 21 trading days before the actual attack), and so on.

It is interesting to note that all the bootstrap tests are not significant, implying that we cannot

RI
reject the null hypothesis of no change in the volatility during the 30 days immediately before the
terrorist attack. In contrast, six out of the eight parametric tests are significant, indicating a

SC
change in the volatility prior to the actual attack. Given that the events we are investigating are
completely unpredictable, there is no explanation for this misleading result, except for problems

NU
with the test itself. Knowing that at least two conditions of the theoretical tests are violated
(normality and serial correlation), we conclude that the bootstrap tests outperform the parametric
tests. Moreover, the considerable difference between the two test statistics indicates that the
MA
reported results of the theoretical tests are biased.

6. Discussion
D

6.1 The Event Study Methodology and the Impact of Terrorist Attacks on Stock Markets
TE

It is expected that terrorist attacks in MENA countries will lead to increased volatility in their
stock markets. It is not obvious, however, for how long the effect will last. Similar questions
P

have been addressed in developed markets. For instance, as outlined in the literature review
section, Essaddam and Karagianis (2014) examined the impact of 44 terrorist attacks on the
CE

stock return volatility of US firms operating worldwide and concluded that the impact lasted for
two weeks. Essaddam and Mnasri (2015) who investigated the impact of 28 terrorist attacks that
AC

happened in 17 countries have drawn a similar conclusion. Our work differs in two aspects. First,
we investigate the impact of terrorist attacks on the volatility of market indices instead of firms‟
return volatility. Second, we address the impact of terrorist attacks on emerging financial
markets that are geographically and culturally related, rather than markets that are not necessarily
similar. Our sample of emerging markets is characterized by an institutional landscape that is
completely different from most markets investigated so far in the literature. The lack of
institutional development seems to lead to effects that last longer in the emerging markets.
Compared with the study of Białkowski et al. (2008), despite the fact that our methodology
benefited from their work, our paper differs in two ways. First, the nature of the events is
completely different, as we consider unpredicted events (terrorist attacks), whereas they
considered predicted events (elections). The second difference consists in our elaboration the
bootstrapping methodology and proving that it outperforms theoretical tests.

24
 ACCEPTED MANUSCRIPT

6.2 The Advantage of Our Bootstrap Method

As outlined in our methodology section, our event study approach benefited from the work of
Białkowski et al. (2008). However, our evaluation approach and bootstrapping algorithm are
quite different from those used in Białkowski et al. (2008). In fact, in their paper, the

PT
bootstrapping consists of randomly drawing (with replacement) N country/date combinations
from the entire set of available countries and dates, to match the number of events. They then
computed the CAV for the randomly generated sample over the respective event window.

RI
Obviously, the null hypothesis might be violated for two reasons. First, there is nothing to
prevent the random process from selecting an estimation period (benchmark) that does not

SC
include the event(s). In this case, the GARCH model cannot serve as a benchmark to forecast
future volatilities under the null hypothesis. Secondly and following the same logic, we might

NU
also have events occurring just after the estimation window (the estimation period). In this case,
even if the estimation period does not include any event, the actual residuals, , used in the
computation of the bootstrapped CAV will not be consistent with the conditional forecasted
MA
volatilities, .

Clearly, the empirical distribution obtained from the bootstrap is not going to be consistent with
the null hypothesis. However, in our paper, the residuals used to compute the empirical
D

bootstrapped CAV distribution were obtained from the original benchmark GARCH model (the
TE

event-free periods). Moreover, these residuals are, by construction, consistent with the
conditional forecasted volatilities , as they are drawn from a distribution with variances that are
equal to the conditional forecasted volatilities. Hence, the bootstrapping technique used in this
P

paper is perfectly consistent with the null hypothesis.

CE

Moreover, in Białkowski et al. (2008), in each bootstrapping iteration, the CAV is computed
from the estimation of N random GARCH equations. Hence, the distribution of countries in each
AC

bootstrapping iteration is different and obviously does not coincide with the distribution of
countries in the original sample. Hence, the bootstrapping here might not help to circumvent the
potential problems caused by the cross-sectional dependence or autocorrelation of the regression
residuals.

7. Conclusion
This study contributes to the existent literature on the impact of terrorist attacks on financial
markets in four ways. First, we investigate the impact of these dramatic events on stock markets
in the MENA countries, a subset of emerging markets, which, to the authors‟ knowledge, was
not subject to this analysis. In fact, the frequency of terrorist attacks is quite high in the MENA
region, which may heighten the fear of investors and affects long-run investment decisions. From
policymakers‟ perspectives, it is very important to know how financial markets react to terrorist
attacks in order to design the appropriate policies. Our second contribution is related to the
dimension of the stock market we are interested in, namely the market volatility. In fact, most of

25
 ACCEPTED MANUSCRIPT

the previous studies have focused on the impact of terrorist attacks on stock market returns and
very few investigate the impact of events other than terrorist attacks (such as wars, elections,
natural disasters, etc.) on stock market volatility. In this paper, we investigate the size effect and
term effect of terrorist attacks on volatility of stock returns in the MENA countries using the
event study approach. Our third contribution consists in the originality of the method we use to

PT
test for the financial integration in the MENA region. In fact, we select a major terrorist attack
that occurred in one of the MENA countries, and we use the event study model to investigate the

RI
impact of this turning event on the financial markets in other MENA countries. Based on the
significance of this impact, we can then conclude about the degree of financial integration

SC
between those markets. The fourth contribution consists in the bootstrapping technique we use to
check the robustness of our findings. In fact, in order to track down the effects of terrorist attacks
on the volatility of financial markets, we evaluate our event study methodology using two

NU
different testing methods. The first test is parametric and conducted on the hypothesis that the
underlying assumptions relative to the calculations of the theoretical p-values are satisfied,
whereas the second test is based on an improved bootstrapping technique that overcomes the
MA
drawbacks of the parametric tests and the traditional bootstrapping techniques. Although the
results of both tests lead to the same conclusion that terrorist attacks have a significant positive
impact on the volatility of financial markets, the semi-parametric test is proven to be more
D

robust, and provides a better approximation of the empirical distribution of test statistics in the
context of volatility event models. Moreover, our results prove that theoretical p-values might be
TE

misleading if underlying assumptions are violated.

P

Our findings have important implications for policymaking. First, we present evidence that
terrorism risk is an important factor in explaining the volatility of stock returns and should be
CE

dealt with if we want more stable financial markets. The fact that the impact of terrorist attacks
lasts for a longer period in emerging markets urges policymakers to enforce financial and
AC

economic institutions, if they want to keep up with the flow of investors. Second, the evidence of
the regional financial integration of MENA markets suggests that the risk of terrorism is a threat
not only to the stability of local economies but also cross-border economies. A lack of
appropriate policies may lead to more social and economic tensions in a region that is already
subject to unrest in many aspects.

26
 ACCEPTED MANUSCRIPT

References
Abadie, A., Gardeazabal, J., 2008. Terrorism and the world economy European Economic
Review 52, 1–27

PT
Abadie, A., Gardeazabal, J., 2003. The economic cost of conflict: A case study of the Basque
country. American Economic Review 93, 113–132.

RI
Abul Basher, S., Nechi, S., Zhu, H., 2014. Dependence patterns across Gulf Arab stock markets:
A copula approach. Journal of Multinational Financial Management 25-26, 30-50.

SC
Aloui, C., Nguyen, D. K., 2014. On the detection of extreme movements and persistent
behaviour in Mediterranean stock markets: a wavelet-based approach. Applied Economics

NU
46 (22), 2611–2622.

Al-Deehani, T., Moosa, I.A., 2006. Volatility spillover in regional emerging stock markets: A
MA
structural time series approach. Emerging Markets Finance and Trade 42, 78–89.

Al-Ississ M. M. (2010). The Cross-border Financial Impact of Violence. Working paper, Belfer
Center for Science and International Affairs, Harvard Kennedy School of Government.
D

Amihud, Y., Wohl A., 2004. Political news and stock prices: the case of Saddam Hussein
TE

contracts. Journal of Banking and Finance 28, 1185‐1200.

Apergis, E., Apergis, N., 2016. The 11/13 Paris terrorist attacks and stock prices: The case of the
P

international defense industry. Finance research letters, forthcoming.

CE

Arin, KP., Ciferri, D., Spagnolo, N., 2008. The price of terror: The effects of terrorism on stock
market returns and volatility. Economics Letters 101 (3), 164- 167.
AC

Aslam, F. Kang, H.-G., Mohti, W., Rafique, A., Salman, A., 2016. The Impact Of Terrorism On
Financial Markets: Evidence From Asia. The Singapore Economic Review, forthcoming

Aslam, F., Kang, H-G., 2015. How Different Terrorist Attacks Affect Stock Markets. Defence
and Peace Economics 26 (6), 634-648.

Assaf, A., 2003. Transmission of stock price movements: the case of GCC stock markets.
Review of Middle East Economics and Finance 1,171–189.

Athanassiou, E., Kollias, C., Syriopoulos, Th., 2006. Dynamic volatility and external security
related shocks: the case of the Athens stock exchange. Journal of International Financial
Markets, Institutions and Money 16, 411- 424.

Bai, J., Ng, S., 2005. Tests for skewness, kurtosis, and normality for time series data. Journal of
Business Economics and Statistics 23, 49–60.

27
 ACCEPTED MANUSCRIPT

Baumert, T., Buesa, M., Lynch, T., 2013. The impact of terrorism on stock markets: the Boston
bombing experience in comparison with previous terrorist events. Working paper, available
at: http://eprints.ucm.es/31249/1/DT%20IAIF%202013%20(88).pdf

Bentes , S. R., 2016. Long memory volatility of gold price returns: How strong is the evidence

PT
from distinct economic cycles? Physica A 443, 149-160.

Beran, R. 1986. Simulated power functions. Annals of Statistics 14, 151–173.

RI
Bialkowski, J., Gottschalk, K., Wisniewski, T., 2008. Stock market volatility around national

SC
elections. Journal of Banking and Finance 32, 1941‐1953.

Boehmer, E., Musumeci, J., Poulsen, A.B., 1991. Event-study methodology under conditions of

NU
event-induced variance. Journal of Financial Economics 30, 253–272.

Bollerslev, T. (1987). A conditionally heteroskedastic time-series model for security prices and
rates of return data. Review of Economics and Statistics 69, 542–547.
MA
Bouri, E., Azzi, G., 2014. On the Dynamic Transmission of Mean and Volatility across the Arab
Stock Markets. Journal of Emerging Market Finance 13 (3), 279–304.
D

Bouri, E., Yahchouchi, G., 2014. Do return and volatility traverse the Middle Eastern and North
TE

African (MENA) stock markets borders? Journal of Economic Studies 41 (2), 317-344.

Brounrn, D., Derwall, J., 2010. The Impact of Terrorist Attacks on International Stock Markets.
P

European Financial Management, 16 (4), 585–598.

CE

Brown, S.J., Warner, J.B., 1985. Using daily stock returns: The case of event studies. Journal of
Financial Economics 14, 3–32.
AC

Carpentier, C., Suret, J-M, 2015. Stock market and deterrence effect: A mid-run analysis of
major environmental and non-environmental accidents. Journal of Environmental
Economics and Management 71, 1–18.

Chen, A.H. and T.F. Siems, T.F., 2004. The effects of terrorism on global capital markets.
European Journal of Political Economy 20 (2), 349–366.

Chesney, M., Ganna Reshetar, G., Karaman, M., 2011. The impact of terrorism on financial
markets: An empirical study. Journal of Banking & Finance 35, 253–267

Choudhry, T., 2005. September 11 and Time-Varying Beta of United States Companies.
Applied Financial Economics 15 (17), 1227-1242.

Choudhry, T., 1995. Stock return volatility and World War II: evidence from GARCH and
GARGH-X models. International Journal of Finance Economics 2, 17-28.

28
 ACCEPTED MANUSCRIPT

Chau, F., Deesomsak, R., Wang, J., 2014. Political uncertainty and stock market volatility in the
Middle East and North African (MENA) countries . Journal of International Financial Markets,
Institutions and Money 28, 1-19.

Davidson R., MacKinnon, J. G. 1999. The size distortion of bootstrap tests. Econometric Theory

PT
15, 361–376.

Drakos, K., 2010. The determinants of terrorist shocks' cross-market transmission. Journal of

RI
Risk Finance, Emerald Group Publishing 11 (2), 147-163.

SC
Drakos, K., 2004. Terrorism-induced structural shifts in financial risk: airline stocks in the
aftermath of the September 11th terror attacks. European Journal of Political Economy, 20,
435-446.

NU
Efron, B., 1979. Bootstrap methods: Another look at the jackknife. Annals of Statistics 7, 1-26.

Eldomiaty, T. I., Al Qassemi, T. B-F, Mabrouk, A. F., Abdelghany, L. S., 2015. Institutional
MA
quality, economic freedom and stock market volatility in the MENA region,
Macroeconomics and Finance in Emerging Market Economies, DOI:
10.1080/17520843.2015.1093011.
D

Essaddam, N., Karagianis, J.M., 2014. Terrorism, country attributes, and the volatility of stock
TE

returns. Research in International Business and Finance 31, 87-100.

Essaddam, N., Mnasri, A., 2015. Event-study volatility and bootstrapping: an international study.
P

Applied Economics Letters 22 (3), 209-2013.

CE

Estrada, M. A. R., Koutronas, E., 2016. Terrorist attack assessment: Paris November 2015
and Brussels March 2016. Journal of Policy Modeling, forthcoming.
AC

Fiordelisi, F., Galloppo, G., Ricci, O., 2014. The effect of monetary policy interventions on
interbank markets, equity indices and G-SIFIs during financial crisis. Journal of
Financial Stability 11, 49-61.

Frey, B., Kucher, M., 2000. World War II as reflected on capital markets. Economics Letters 69,
187‐191.

Frey, B., Kucher, M., 2001. Wars and markets: How bond values reflect the Second World War.
Economica 68, 317‐333.

Guidolin, M., La Ferrara, E., 2010. The economic effects of violent conflict: evidence from asset
market reactions. Journal of Peace Research, 47, 671‐684.

Günster, A., van Dijk, M., 2016. The impact of European antitrust policy: Evidence from the
stockmarket. International Review of Law and Economics 46, 20–33.

29
 ACCEPTED MANUSCRIPT

Haitsma, R., Unalmis, D., de Haana, J., 2016. The impact of the ECB‟s conventional and
unconventional monetary policies on stock markets. Journal of Macroeconomics 48, 101–
116.

Hall, P., 1992. Effect of bias estimation on coverage accuracy of bootstrap confidence intervals

PT
for a probability density. Annals of Statistics 20, 675-694.

Hall, P., Horowitz, J.L., 1996. Bootstrap critical values for tests based on generalized-method-

RI
of moments estimators. Econometrica 64, 891-916.

SC
Hall, P., Titterington D.M., 1989. The effect of simulation order on level accuracy and power of
Monte-Carlo tests. Journal of the Royal Statistical Society Series B 51, 459–467.

NU
Hilliard, J.E., Savickas, R., 2002. On the statistical significance of event effects on unsystematic
volatility. Journal of Financial Research 25, 447–462.

Hobbs, J., Ludwig Christian Schaupp, Joel Gingrich , (2016),"Terrorism, militarism, and stock
MA
returns", Journal of Financial Crime 23 ( 1), 70-86.

Hon, M., Strauss, J., Young, S.-K., 2004. Contagion in financial markets after September 11:
D

myth or reality. The Journal of Financial Research 27 (1), 95-114.

TE

Horowitz, J.L., 1994. Bootstrap-based critical values for the information-matrix test. Journal of
Econometrics 61, 395-411.
P

Hwang, S., Valls Pereira, P.L., 2006. Small sample properties of GARCH estimates and
CE

persistence. European Journal of Finance 12, 473–494.

Imbierowicz, B., Mark Wahrenburg, M., 2013. Wealth transfer effects between stockholders and
AC

bondholders. The Quarterly Review of Economics and Finance 53, 23–43.

Johnston, R.B. and Nedelescu, O.M., 2005. The Impact of Terrorism on Financial Markets. IMF
Working Paper WP/05/60.

Karolyi, A, Martell, R., 2010. Terrorism and the Stock Market. International Review of Applied
Finance Issues and. Economics 2 (2), 285-314.

Kenjegaliev, A., Duygun, M., Mamedshakhova, D., 2016. Do rating grades convey important
information: German evidence? Economic Modelling 53, 334–344.

Kolaric, S., Schiereck, D., 2016. Are stock markets efficient in the face of fear? Evidence from
the terrorist attacks in Paris and Brussels. Finance Research Letters, forthcoming.
Kollias, C., Papadamou, S., Stagiannis, A., 2011. Terrorism and capital markets: The effects of
the Madrid and London bomb attacks. International Review of Economics & Finance 20 (4),
532-541.

30
 ACCEPTED MANUSCRIPT

Kollias, C., Papadamou, S., Stagiannis, A., 2010. Armed Conflicts and Capital Markets: the Case
of the Israeli Military Offensive in the Gaza Strip. Defence and Peace Economics 21,
357‐365.

Kothari, S.P., Warner, J.B., 2006. Econometrics of event studies. in B Espen Eckbo (ed)

PT
Handbook of Corporate Finance: Empirical Corporate Finance Vol. A (North Holland,
Elsevier, 2006)

RI
Ljung, G. M., Box, G. E. P., 1978. On a measure of lack of fit in time series models, Biometrika
65, 297-303.

SC
Marashdeh, H., 2005. Stock Market Integration in the MENA Region: An Application of the
ARDL Bounds Testing Approach, Working Paper 05-27, Department of Economics,

NU
University of Wollongong.

Neaime, S., 2016. Financial crises and contagion vulnerability of MENA stock markets.
MA
Emerging Markets Review, forthcoming.

Nikkinen, J., Vähämaa, S., 2010. Terrorism and Stock Market Sentiment. Financial Review 45
(2), 263-275.
D

Schneider, G., Troeger, V., 2006. War and the world economy. Stock market reactions to
TE

international conflicts. Journal of Conflict Resolution 50, 623‐645.

Scholtens, B., Boersen, A. 2011. Stocks and energy shocks: The impact of energy accidents on
P

stock market value. Energy 36, 1698-1702.

CE

Smales, L. A., 2014. Non-scheduled news arrival and high-frequency stock market dynamics
Evidence from the Australian Securities Exchange. Research in International Business and
AC

Finance 32, 122–138.

Wang, S., Mayes, D. G., 2012. Monetary policy announcements and stock reactions: An
international comparison. North American Journal of Economics and Finance 23, 145–164.

Yu, J-S., M., Hassan, M.K, 2008. Global and regional integration of the Middle East and North
African (MENA) stock markets. The Quarterly Review of Economics and Finance 48 (2008)
482–504.

31